[rwgk]

Highlights

Separate three CUDA 12 runtime backports from the source/build integration in #3173. Each fix has its own commit; the existing focused regression tests move with the relevant fixes. This keeps the migration review focused while preserving the previously integrated behavior.

• Accept byte-string compiler options alongside strings, retaining the nvJitLink regression and NVVM bytes-options case.
• Retain borrowed coredump setter buffers and release getter-owned storage.
• Preserve exceptions raised inside registered handle getters instead of misreporting an unknown handle type.

Provenance and Review Scope

These changes were already present in the staged #2737 implementation; this PR relocates them without further runtime changes.

Fix	Earlier upstream change	Coverage retained here
Byte-string compiler options	#1665, #1698	nvJitLink bytes-options test and NVVM parameterized case
Coredump buffer ownership	#2399	Existing API tests remain; ordinary round-trip tests do not detect the getter leak
Registered handle-getter errors	#2551	Focused regression proving an internal KeyError propagates

The coredump change has no new memory-instrumentation test. Upstream removed an ineffective lifetime regression test; a normal functional round trip would also pass with the getter leak. This extraction preserves that validation boundary.

Runner-specific NVML test maintenance belongs to #3176. The Windows constant declaration remains in #3173 because its Cython warning would otherwise fail the newly enabled warnings-as-errors build.

Stack / Review Boundary

Layer 8 of 8 in native stack #3179, based on #3178. The stack is rebased onto main dfad7e3658a9. This layer's diff and the complete CUDA 12 source tree are unchanged by the rebase; #3178 records current local checks and earlier integrated QA.

Validation

Full hosted CI passes at #3182 for this exact head, 785641713c4d, on its first attempt: all 28 wheel builds, the CUDA 12/CUDA 13 GPU matrix, source-distribution checks, and documentation checks passed. Full hosted CI passes at #3176 for the earlier merge group.

At this head, 436 CI-tool tests and 55 subtests pass, together with full local pre-commit. #3178 removes the active legacy backport setting while retaining compatibility with earlier schema-2 tagged registries. #3176 bounds the cuDSS test dependencies to the loader's supported ABI.

• Full local pre-commit run --all-files --show-diff-on-failure passes, including Lychee.
• Comparison with the pre-rebase head confirms that this layer's diff and the complete CUDA 12 source tree are unchanged.
• The narrowed cuda.bindings: integrate CUDA 12 source with shared build tooling #3173 also passes full local pre-commit; its earlier build-helper suite passed with 75 passed and 3 skipped.

Checklist

• Existing regression tests move with the corresponding runtime changes.
• The description records provenance, scope, and validation limits.

[copy-pr-bot]

Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually.

Contributors can view more details about this message here.

[rwgk]

/ok to test 7856417

[github-actions]

Doc Preview CI
🚀 View preview at https://nvidia.github.io/cuda-python/pr-preview/pr-3182/
https://nvidia.github.io/cuda-python/pr-preview/pr-3182/cuda-core/
https://nvidia.github.io/cuda-python/pr-preview/pr-3182/cuda-bindings/
https://nvidia.github.io/cuda-python/pr-preview/pr-3182/cuda-pathfinder/
Preview will be ready when the GitHub Pages deployment is complete.